diff --git a/docs/how-to/export-dataset-snapshot.md b/docs/how-to/export-dataset-snapshot.md
index 68b0e38b..c15cf434 100644
--- a/docs/how-to/export-dataset-snapshot.md
+++ b/docs/how-to/export-dataset-snapshot.md
@@ -63,9 +63,9 @@ media paths are relative to the export directory. The JSON marker is written
 last and the completed directory is activated atomically.
 
 Under the same format version `1`, an additive `integrity` block records
-delivery integrity so a later verifier (not shipped here) can tell whether the
-published bytes are still intact. The original `tables` map is unchanged for
-external readers; receipts live only under `integrity`:
+delivery integrity so `hflow verify snapshot` (or `verify_dataset_snapshot`)
+can tell whether the published bytes are still intact. The original `tables`
+map is unchanged for external readers; receipts live only under `integrity`:
 
 | Field | Meaning |
 | --- | --- |
diff --git a/src/hflow/snapshot.py b/src/hflow/snapshot.py
index d005408e..e05716d4 100644
--- a/src/hflow/snapshot.py
+++ b/src/hflow/snapshot.py
@@ -248,10 +248,10 @@ def _build_snapshot_integrity_marker_fields(
 
     Required Parquet tables and every regular file under ``assets/`` (copy mode)
     get ``path`` / ``size_bytes`` / ``sha256``. ``content_id`` digests that
-    normalized inventory so a deleted member is detectable without a verifier
-    product yet. References mode leaves ``assets`` empty: remote media are not
-    fetched for hashing. Copy mode re-reads each copied asset once after the
-    copy to compute its hash.
+    normalized inventory so ``verify_dataset_snapshot`` can detect a deleted
+    member even when every remaining file still matches. References mode leaves
+    ``assets`` empty: remote media are not fetched for hashing. Copy mode
+    re-reads each copied asset once after the copy to compute its hash.
     """
     tables: dict[str, FileIntegrityRecord] = {}
     for table_name, file_name in _REQUIRED_TABLE_FILES.items():